Skip to content

resnet9

ResNet9

Bases: Module

A 9-layer ResNet PyTorch model for cifar10 image classification. The model architecture is from https://github.com/davidcpage/cifar10-fast

Parameters:

Name Type Description Default
input_size Tuple[int, int, int]

The size of the input tensor (channels, height, width). Both width and height of input_size should not be smaller than 16.

 (3, 32, 32)
classes int

The number of outputs.

 10

Raises:

Type Description
ValueError

Length of input_size is not 3.
ValueError

input_size[1] or input_size[2] is not a multiple of 16.
Source code in fastestimator/fastestimator/architecture/pytorch/resnet9.py 
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
class ResNet9(nn.Module):
    """A 9-layer ResNet PyTorch model for cifar10 image classification.
    The model architecture is from https://github.com/davidcpage/cifar10-fast

    Args:
        input_size: The size of the input tensor (channels, height, width). Both width and height of input_size should
            not be smaller than 16.
        classes: The number of outputs.

    Raises:
        ValueError: Length of `input_size` is not 3.
        ValueError: `input_size`[1] or `input_size`[2] is not a multiple of 16.
    """
    def __init__(self, input_size: Tuple[int, int, int] = (3, 32, 32), classes: int = 10):
        ResNet9._check_input_size(input_size)
        super().__init__()
        self.conv0 = nn.Conv2d(input_size[0], 64, 3, padding=(1, 1))
        self.conv0_bn = nn.BatchNorm2d(64, momentum=0.2)
        self.conv1 = nn.Conv2d(64, 128, 3, padding=(1, 1))
        self.conv1_bn = nn.BatchNorm2d(128, momentum=0.2)
        self.residual1 = Residual(128)
        self.conv2 = nn.Conv2d(128, 256, 3, padding=(1, 1))
        self.conv2_bn = nn.BatchNorm2d(256, momentum=0.2)
        self.residual2 = Residual(256)
        self.conv3 = nn.Conv2d(256, 512, 3, padding=(1, 1))
        self.conv3_bn = nn.BatchNorm2d(512, momentum=0.2)
        self.residual3 = Residual(512)
        self.fc1 = nn.Linear(512, classes)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        # prep layer
        x = self.conv0(x)
        x = self.conv0_bn(x)
        x = fn.leaky_relu(x, negative_slope=0.1)
        # layer 1
        x = self.conv1(x)
        x = fn.max_pool2d(x, 2)
        x = self.conv1_bn(x)
        x = fn.leaky_relu(x, negative_slope=0.1)
        x = x + self.residual1(x)
        # layer 2
        x = self.conv2(x)
        x = fn.max_pool2d(x, 2)
        x = self.conv2_bn(x)
        x = fn.leaky_relu(x, negative_slope=0.1)
        x = x + self.residual2(x)
        # layer 3
        x = self.conv3(x)
        x = fn.max_pool2d(x, 2)
        x = self.conv3_bn(x)
        x = fn.leaky_relu(x, negative_slope=0.1)
        x = x + self.residual3(x)
        # layer 4
        x = nn.AdaptiveMaxPool2d((1, 1))(x)
        x = torch.flatten(x, 1)
        x = self.fc1(x)
        x = fn.softmax(x, dim=-1)
        return x

    @staticmethod
    def _check_input_size(input_size):
        if len(input_size) != 3:
            raise ValueError("Length of `input_size` is not 3 (channel, height, width)")

        _, height, width = input_size

        if height < 16 or width < 16:
            raise ValueError("Both height and width of input_size need to not smaller than 16")

Residual

Bases: Module

A two-layer unit for ResNet9. The output size is the same as input.

Parameters:

Name Type Description Default
channel

Number of input channels.

 required
Source code in fastestimator/fastestimator/architecture/pytorch/resnet9.py 
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
class Residual(nn.Module):
    """A two-layer unit for ResNet9. The output size is the same as input.

    Args:
        channel: Number of input channels.
    """
    def __init__(self, channels: int):
        super().__init__()
        self.conv1 = nn.Conv2d(channels, channels, 3, padding=(1, 1))
        self.conv1_bn = nn.BatchNorm2d(channels)
        self.conv2 = nn.Conv2d(channels, channels, 3, padding=(1, 1))
        self.conv2_bn = nn.BatchNorm2d(channels)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        x = self.conv1(x)
        x = self.conv1_bn(x)
        x = fn.leaky_relu(x, negative_slope=0.1)
        x = self.conv2(x)
        x = self.conv2_bn(x)
        x = fn.leaky_relu(x, negative_slope=0.1)
        return x